💳 Finance Club Project - Credit Default Prediction

Goal: Predict credit card defaults with a focus on Recall (capturing risk) using an F2-Score optimized XGBoost pipeline.

Tip

• Problem Statement: Finclub Summer Project 2 (2025).pdf
• Final Report: Finance_club_ML_project_Report.pdf
• Experimental Notebook: Finance_club_project (2).ipynb

---

📊 Pipeline Architecture

graph TD
    %% Nodes
    A[📂 Raw CSV Data] --> B(🛠️ Feature Engineering)
    B -->|AVG_Bill_amt, Ratios| C{⚙️ Preprocessing}
    
    C -->|Train Split| D[⚖️ SMOTE Resampling]
    C -->|Test Split| E[📏 Scaling]
    
    D --> F[🚀 XGBoost Training]
    F --> G[🎛️ Threshold Tuning]
    E --> G
    
    G -->|Maximize F2| H[📈 Final Evaluation]
    H --> I[📄 Results & Plots]
    
    %% Styling
    style A fill:#f9f,stroke:#333,stroke-width:2px
    style F fill:#ff9,stroke:#333,stroke-width:2px
    style H fill:#9f9,stroke:#333,stroke-width:2px

Loading

---

🚀 Key Features

Feature	Description
Modular Design	Separation of Data, Features, and Models for production readiness.
F2-Score Focus	Optimized specifically to catch defaulters (High Recall) over simple accuracy.
Smart Tuning	Threshold optimization loop (0.01 to 1.0) to find the perfect cut-off.
Visual Artifacts	Automatically generates ROC, PR Curves, and Confusion Matrices per run.

---

📂 Project Structure

FinanceClub_Project/
├── 📄 main.py                  # 🚀 Pipeline Entry Point
├── 📂 config/
│   └── config.yaml             # ⚙️ Hyperparameters & Paths
├── 📂 data/
│   └── raw/                    # 📥 Input: train_dataset_final1.csv
├── 📂 logs/                    # 📝 Execution Logs
├── 📂 results/                 # 📊 Outputs (Metrics, Plots, Predictions)
│   └── run_20231228_.../       #    Timestamped Artifacts
├── 📂 src/
│   ├── 🛠️ features/            #    Engineering, Preprocessing, Resampling
│   ├── 🤖 models/              #    XGBoost Trainer, Tuner
│   └── 🔌 data/                #    Loaders
└── 📦 requirements.txt         #    Dependencies

---

⚡ Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Run Pipeline

python main.py

3. View Results

Checking results/ will show:

• metrics.json: Accuracy, Precision, Recall, F2, AUC.
• roc_curve.png: Model discrimination performance.
• feature_importance.png: What drives the predictions?

---

🧠 Model Details

Note

The model uses XGBoost with a heavily weighted positive class (scale_pos_weight: ~6.38) to address the 80/20 class imbalance.

• Algorithm: XGBoost Classifier
• Best Params: n_estimators: 300, max_depth: 7, learning_rate: 0.05
• Imbalance Strategy: SMOTE (Synthetic Minority Over-sampling) on Training Data only.

---

Generated for Finance Club Project Refactoring